Crack Detection using Spectral Clustering: Self-Tuning Considering Crack Feature and Connections

Abstract

Cracks on the pavement road cause various traffic problems. Hence, we should repair them properly. Nowadays, there are a variety of crack detection method based on computer vision for operation efficiency. Spectral Clustering is one of them and effective. However, detection accuracy may decrease depending on the image because roads often have many bumps. In this paper, we cluster images having less noise into two clusters, while noisy images into three clusters. The number of clusters is determined automatically by considering the relationship between crack feature and the number of connections.